Panorama technology Overview

 

I. Image Acquisition

There are two ways to obtain panoramic image materials: one is to use a special panoramic device, such as a panoramic camera or a camera with a fish eye or wide angle lens; the other is to use a general camera to take partial images, then, after projection, the system concatenates to form a panorama.

The advantage of the first method is that it is easy to operate without complex modeling and can easily form a panorama. The disadvantage is that dedicated devices are expensive and difficult to popularize and use. The second method has high requirements on shooting. Generally, you need to use some devices, such as a tripod, to complete shooting. Compared with the former, it is more complex, but the cost is low.

Ii. image projection

Because adjacent local scene images are taken from different angles after the camera turns a certain angle, their projection plane has a certain angle. If you splice a local image seamlessly, the visual consistency in the actual scenario will be damaged, such as turning a curve into a straight line, and it is difficult to splice it seamlessly. In order to maintain the Spatial Constraints in actual scenarios, the real-time image taken must be projected onto a certain surface, and the image information must be saved as a curved surface on the computer. After the projection is complete, the rotation relationship is removed, and the translation relationship is retained to prepare for image mosaic. Generally, the most common panoramic projection methods include sphere projection, cylindrical projection, and cube projection.

1. SPHERICAL MODEL

2. Cylindrical Surface Model

3. cube Model

The panorama model provides a 360-degree horizontal view of the scenario, and a 180-degree vertical view of the sphere and cube, the scenario has a very high fidelity.

Iii. Image Stitching

Image Stitching is one of the key technologies of panoramic technology and a key link in panoramic production. Due to the limited camera angle and the high price of panoramic cameras, the Research on splicing technology is very meaningful.

1. Main purposes of splicing

Image Stitching Technology can solve the problem that the camera and other imaging instruments cannot take a large image at a time due to the limited angle of view. It uses a computer for matching and synthesis of a wide-angle image. Therefore, it is widely used in practical use.

2. Image Mosaic Problems

The key to image mosaic is to precisely find the exact position of the overlapping parts of the adjacent two images, then determine the location transformation relationship between the two images, and finally splice and edge fusion. Because the camera is affected by environment and hardware conditions, the images are often different in terms of translation, rotation, scaling, perspective deformation, chromatic aberration, and distortion, these differences greatly increase the difficulty and complexity of image mosaic. In practical application, the most basic splicing technology mainly involves three changes: translation, scaling, and rotation (A is the original image, and A' is the transformed image ):

(1) Translation: a' = a + K;

(2) ZOOM: a' = SA, where S is the zoom ratio;

(3) rotation: a' = Ra, which is the rotation matrix.

In these three changes, scaling and rotation are difficult to solve. How to Determine the scaling ratio and rotation angle between the two images is the difficulty of image mosaic technology.

3. Image Mosaic technical process and core issues

The basic process of image mosaic technology.

Core problem: Image Mosaic effect is closely related to a key technology, that is, image matching.

4. Image Matching

Image matching is designed to search for vertices with the same name. It is one of the core topics of computer vision and digital photography measurement, it refers to finding the overlapping position and range (also called image alignment) of some overlapping sequence images ). The two images may be taken from different time, different environments, different sensors, or different perspectives, which results in not only noise, in addition, there are severe gray and geometric distortion.Image matching is assumed that there are two rectangular areasA and B. It is known that B contains a region A', and A' and a are the same modules. Locate the location of a' in B.

From a macro perspective, image matching methods can be classified into gray-scale matching and feature matching.Grayscale matchingIt is represented by the commonly used correlation coefficient method, which mainly considers the gray distribution and statistical characteristics in the local area;Feature MatchingFirst, we need to extract features, such as edge features, texture features, and information entropy features. Then, we use chaincodes or parameters to represent these features and use them as matching elements. In fact, feature matching is often used in combination with grayscale matching, that is, some feature quantity constraints are introduced during grayscale matching.

Image matching system process.

The algorithms involved in this section still need to be further studied and sorted out.

Iv. Inverse Projection

To reconstruct the view corresponding to each line of sight in the viewpoint space from the panoramic image, the panoramic image processed by the forward projection must be reversed. The Inverse Projection Algorithm of panoramic images solves the following problems: Re-constructs the view corresponding to each line of sight in the spherical viewpoint space from the panoramic images.

V. References

1. panorama technology research, Li yunwei, master's degree thesis from Huazhong University of Science and Technology, 2007.

2. Analysis of panorama technology application status and several technical problems, Hu xuejin, keyuan, November 24, 2009, pp. 283.

3. Design and Implementation of panorama stitching algorithm, article 21st, Journal of Chongqing Institute of Technology (Natural Science Edition), pp. 9th.